Fix autolog write recovery and active-session pruning

[ctoth]

Summary

• keep the serialized AutoLog flush queue usable after an IndexedDB append or prune failure while still rejecting the failed flush to its caller
• always release the current session after a failed final flush so a later session can start
• exclude the active session from whole-session pruning so later entries remain reachable and accounted for

Root cause and impact

AutoLogService stored the rejecting operation itself as the next queue dependency, permanently poisoning every later flush. Its failed endSession() path also skipped session cleanup. Separately, AutoLogStore treated the currently-written session as an ordinary oldest-session deletion candidate, orphaning later entries and defeating the storage cap.

TDD

The new regressions failed first by proving that the queue stayed rejected, the failed session could not be replaced, and the active session record was deleted. The implementation then made those same tests pass.

Validation

• npm test -- src/logging/AutoLogService.test.ts src/logging/AutoLogStore.test.ts — 2 files, 11 tests
• npm test — 106 files, 1,055 tests
• npm run typecheck
• npm run lint:staged — passes with three pre-existing warnings in the touched AutoLog files
• npm run build
• git diff --check

Closes #53
Closes #57